Visual walkthrough — Row echelon form and reduced row echelon form
Everything below uses one running example, the same system from the parent's Worked Example 1:
Step 1 — What is a linear equation, really? (a picture, not a formula)
WHAT. Take just the first equation, . The letters are three unknown numbers we are hunting for. The equation is a rule: whatever the three numbers are, when you take copy of , copies of , and copy of and add them, you must get exactly .
WHY start here. Before we shuffle any rows, we must see what one equation is. A linear equation with three unknowns is not a curve — it is a flat sheet (a plane) floating in the room. Every point on that sheet is one valid triple .
PICTURE. Below, the room's three directions are the , , axes. The single red sheet is all triples obeying . Notice: infinitely many solutions for one equation alone — the freedom shrinks only when we add more rules.
Step 2 — Three equations = three sheets; a solution = where they all meet
WHAT. Our system has three equations, so three planes. A triple solves the whole system only if it lies on all three sheets at once.
WHY. This is the geometric meaning of "solve the system": find the common point(s). For three generic planes the overlap is a single point — one , one , one . That point is the answer we will chase.
PICTURE. Three sheets slicing through the room. The red dot is their unique meeting point. Everything we do from now on is just moving the sheets around without moving that red dot — that is the whole trick.
Step 3 — Package the sheets as a grid of numbers (the augmented matrix)
WHAT. Writing over and over is wasteful — the positions already tell us which unknown a number belongs to. Strip the letters; keep the coefficients and constants in a grid.
WHY this tool and not staying with words. A matrix turns "manipulate three equations" into "manipulate three rows of numbers." The vertical bar is a fence: left of it are coefficients (the sheet's tilt), right of it is the constant (how far the sheet sits from the origin).
PICTURE. Each row is colour-labelled to its plane. The red column, right of the bar, is the constants — the part that says where each sheet floats.
Step 4 — One legal move: subtract a row (why the dot stays put)
WHAT. Take Row 2 and subtract Row 1 from it: . In equation language, equation 2 becomes (equation 2) minus twice (equation 1).
WHY it is legal — the heart of everything. If a point already lies on both sheet 1 and sheet 2, then it satisfies equation 1 and equation 2, so it satisfies any combination of them — including "equation 2 minus twice equation 1." And this move is reversible: add back to recover the original. Reversible means no solution is created and none is destroyed — the red dot cannot move.
PICTURE. The blue sheet (equation 2) tilts into a new orientation, but it still passes through the red dot. That is the entire justification for elimination: we are allowed to swing sheets around their common pin.
Term by term, what the subtraction does to Row 2:
- The leading became — that was the goal: erase from equation 2.
- , , — every other entry updates in lockstep so the sheet still holds the red dot.
Step 5 — Carve the staircase (reach REF)
WHAT. Repeat the "subtract to make zeros below" move, working left to right, top to bottom. From Step 4 we also do , then :
\;\xrightarrow{R_3\to R_3-R_2}\; \left[\begin{array}{ccc|c} \mathbf{1}&2&1&6\\ 0&\mathbf{1}&1&3\\ 0&0&\mathbf{1}&2\end{array}\right]\ \textbf{(REF)}$$ **WHY.** Each column's first non-zero entry (the bold **pivot**) becomes the *only* thing anchoring its variable from that row downward. Row 3 now contains *just* $z$: $z = 2$. That is the point of the staircase — the bottom step traps a single unknown. **PICTURE.** Watch the pivots (red) march **down and to the right**, forming steps. Below each pivot: all zeros. This shape *is* Row Echelon Form. - Row 3 = $(0,0,1\mid 2)$ means $0x+0y+1z = 2$, i.e. $z=2$. **One variable, solved.** - Row 2 = $(0,1,1\mid 3)$ means $y + z = 3$; knowing $z=2$ gives $y=1$ — this is ==back-substitution==. - Row 1 gives $x + 2y + z = 6 \Rightarrow x = 2$. Climb the stairs, fill in as you go. > [!definition] Pivot and REF > The ==pivot== is the first non-zero entry of a row. A matrix is in **REF** when zero rows sit at the bottom, each pivot lies strictly right of the one above (the staircase), and everything **below** each pivot is $0$. REF is *not* unique — many staircases (different pivot values) reach the same dot. Rank $=$ number of pivots; see [[Rank of a Matrix]]. --- ## Step 6 — Clean above the pivots (reach RREF, read the answer off) **WHAT.** Now go **bottom-up**: scale each pivot to $1$ (already $1$ here) and subtract to kill entries **above** each pivot. $$\left[\begin{array}{ccc|c} 1&2&1&6\\ 0&1&1&3\\ 0&0&1&2\end{array}\right] \xrightarrow[R_1\to R_1-R_3]{R_2\to R_2-R_3} \left[\begin{array}{ccc|c} 1&2&0&4\\ 0&1&0&1\\ 0&0&1&2\end{array}\right] \xrightarrow{R_1\to R_1-2R_2} \left[\begin{array}{ccc|c} \mathbf 1&0&0&2\\ 0&\mathbf 1&0&1\\ 0&0&\mathbf 1&2\end{array}\right]\ \textbf{(RREF)}$$ **WHY.** When each pivot is the **only** non-zero in its column, every row degenerates to $x = \text{number}$. No climbing, no back-substitution — the constants column *is* the answer. This is [[Gauss-Jordan Elimination]]. **PICTURE.** The sheets have been tilted until they align with the coordinate walls: one sheet is "$x=2$", one is "$y=1$", one is "$z=2$". Three perpendicular walls crossing at the red dot — the answer literally readable off the axes. Reading the last grid row by row: $x=2,\ y=1,\ z=2.$ ✅ (Same red dot from Step 2 — it never moved.) --- ## Step 7 — Why RREF is *unique* (the missing piece) **WHAT.** REF was not unique; RREF is. Why? **WHY.** RREF pins down *both* the pivot **positions** and the pivot **column shapes**. The pivot positions are forced: the leftmost column that isn't fully zero across the solution set *must* hold a pivot — there is no choice about *where* steps go, only about the numbers, and RREF fixes the numbers too (pivot $=1$, zeros above and below). The freedom that made REF non-unique (any non-zero pivot value, any leftover entries above) is *exactly* what conditions 4–5 remove. Two people, two different move-sequences, one destination. **PICTURE.** Two different staircases (two REFs) of the *same* matrix — different pivot values, different upper entries — but when each is fully cleaned, both collapse onto the *identical* RREF grid. > [!definition] RREF (and its uniqueness) > **RREF** = REF plus: every pivot is $1$, and each pivot is the *only* non-zero entry in its column. For any matrix there is exactly ==one== RREF. This is why RREF can *define* rank, free variables ([[Free and Basic Variables]]), and even the [[Matrix Inverse via RREF|inverse]] unambiguously. --- ## Step 8 — The degenerate cases (never skip these) **WHAT & WHY.** Not every system has one dot. Two things can go "wrong," and the staircase *tells you which*: **Case A — a free variable (a whole line of solutions).** If there are fewer pivots than unknowns, a column has no pivot; that variable is **free**. Example from the parent: $$\left[\begin{array}{ccc|c} 1&2&0&1\\ 0&0&1&1\end{array}\right]\Rightarrow x=1-2y,\ z=1,\ y\text{ free.}$$ The two planes meet not in a dot but along a *line* — one degree of freedom left over. **Case B — inconsistency (no solution).** If a pivot lands in the **constants** column: $$\left[\begin{array}{cc|c} 1&1&2\\ 0&0&1\end{array}\right]\Rightarrow 0=1.$$ Row 2 says $0x+0y = 1$: an impossible sheet, the planes are parallel and never share a point. Empty solution set. **PICTURE.** Left: two sheets crossing in a red line (free variable). Right: two parallel sheets with a gap — no red dot anywhere (inconsistent). > [!mistake] The pivot-in-the-last-column trap > A pivot in the constants column *feels* like an ordinary pivot, but it decodes to $0 = \text{nonzero}$. Always read the last column's pivots as a **contradiction alarm**, not a variable. --- ## The one-picture summary One frame, the whole journey: messy sheets (Step 2) → subtract to zero-out below (Steps 4–5, the red staircase forms) → clean above (Step 6, sheets snap to the axis-walls) → read the red dot. The bottom strip lists the two escape hatches: a leftover line (free variable) or parallel gap (inconsistent). > [!recall]- Feynman: the walkthrough in plain words > Picture three big glass sheets floating in a room; the one spot where all three overlap is the secret answer. Solving is allowed to *tilt* the sheets, but only in ways that keep them pinned through that same overlap spot — because "subtract one equation from another" can always be undone, so nothing is lost. We tilt them into a neat staircase where the bottom sheet mentions only one letter (so it hands us that letter for free), then we climb up filling in the rest. If we keep polishing until each sheet lines up flat against a wall — one sheet says "$x$ is here", one "$y$ is here", one "$z$ is here" — we don't even climb; the answer is written on the walls. That perfectly-polished picture is always the same no matter how you got there — that's why RREF is unique. And if the sheets meet in a whole line instead of a dot, some letter is free to roam; if two sheets are parallel and never touch, there's simply no answer at all. --- ## Connections - [[Gaussian Elimination]] — carves the staircase (Steps 4–5). - [[Gauss-Jordan Elimination]] — cleans above the pivots (Step 6). - [[Rank of a Matrix]] — the number of pivots you carved. - [[Solving Systems of Linear Equations]] — the geometric meaning of Steps 1–2. - [[Free and Basic Variables]] — Step 8, Case A. - [[Matrix Inverse via RREF]] — RREF's uniqueness put to work. - [[Linear Independence]] — pivot columns are the independent directions. - [[Elementary Matrices]] — each legal move as a matrix on the left.